rpast.bsky.social

TP3 is a pinnacle of storytelling.

New prompt to try out in midjourney

Ever wanted to do deep learning with a neural net that is one-to-one mapped to 65.05% of the fruit fly brain? 😄 Before this year ends, I will quickly advertise our code release of `flyvis.` Docs: t.ly/YqWzR Repo: t.ly/pMWpp Work with @jakhmack.bsky.social, @srinituraga.bsky.social and colleagues

Mars Express is the most cyberpunk animation I've seen since Ghost in the Shell. I'm amazed by the attention to detail, the depth of the characters, and the supremacy of style above all else—not to mention the stunning animation.

Statistical analysis, archaeology, class struggle, mysterious Ismaili games from India: Board Game Studies is easily my all-time favorite journal, tied only with Lingua Aegyptia. Finally getting around to reading Vol 18 from Oct ‘24. sciendo.com/issue/BGS/18/1

I like the way the Christian author of *The Cloud of Unknowing* describes contemplative prayer: "Come to rest in a naked, blind feeling of your own being, and let go of the conceit that it's yours."* _____ * "conceit" is how you spelled "concept" in Middle English

You will be visited by three ghosts

And that night when the Holy Spirit descends upon people

youtu.be/2adrHe8ntcg?... With all the love

Yay!

Science slowly catching up with an idea that emerged 25 hundreds yrs ago in Gangetic Plain.

Observing model's convergence is like looking at fire

Great sources re baroque music

Mona Lisa overdrive vibes

good day youtu.be/GP6HCykNy3A?...

It's all little gnomes pulling levers behind the curtain

I will soon delete all my twitter posts, so in the next few days I will be recycling some of my most liked tweets, in random order. The first one had to be a shitpost, of course

s1 was a good idea badly implemented. it looks like its not going to change in s2.

Monitoring for dummies 1. Basic stuff - How many people use it - How fast it answers - If it crashes 2. What goes in - Check what users ask - Look for bad requests - Watch for attacks 3. What comes out - Are answers good? - Do they make sense? - Are they safe?

punching above your weight is a useful reframing for impostor syndrome

attention is all you need

looks great! would be even better if all data stayed on localhost though

Can't speak to this instance, but it's frequent enough that when working with time series data, I reindex w/ expected frequency. Ex: ``` full_range = pd.date_range(df['day'].min(), df['day'].max(), freq='D') df = df.set_index('day').reindex(full_range).reset_index() ``` Then search for nulls

My thought for this Tuesday ☕️☕️☕️

tell the ai compliance team in the bank you are going to use a bandit algorithm to optimize the model

what a trip...

This is such an interesting project!:

We all know what 3D printing is, and 2D printing is obviously paper, which means print("Hello world") is 1D printing.

This.

OK, so maybe what Kevin Roose bungled here was that one of the paper's authors (Polosukhin) had made a comparison to "Arrival" ("believed self-attention was a bit like ...") - but without claiming it had been the actual inspiration www.ft.com/content/37bb...

A few things I'm thinking about: - Gaps in AGI safety and policy - Raising literacy on AI risks and impacts through storytelling - Limitations of evals - AI moral patienthood - AI and epistemics - Mechanism design - Metascience Let me know if you have thoughts on these!

disturbing

You're misunderstanding the space of probability at play in Bayesian reasoning. It's epistemic not objective. If I'm on gameshow & prize is behind one of three doors, I assign a probability of 1 in 3 that it's behind door A, even though the prize is objectively speaking 100% behind a single door.

Fuck it! Structured Generation w/ SmolLM2 running in browser & WebGPU 🔥 Powered by MLC Web-LLM & XGrammar ⚡ Define a JSON schema, Input free text, get structured data right in your browser - profit!!

my daughter asked me, 'what is faster than time?' and i haven't been able to stop thinking about it

this is going to be hell of a stress test

FYI, here's the entire code to create a dataset of every single bsky message in real time: ``` from atproto import * def f(m): print(m.header, parse_subscribe_repos_message()) FirehoseSubscribeReposClient().start(f) ```

The myth of scale, in particular, has sunk in deeply: many people now believe that more data is always better as long as it's authentic and human — never mind how inane! Honestly, I think this misunderstanding survives largely because people are flattered by the idea their every word has value

application of social engineering to narrow down interaction data to a meaningful sample of agents. brilliant.

I’m glad you asked.. It’s just a manually curated list but made interacting with this topic much more pleasant already. bsky.app/profile/did:...

this is interesting. an open-source engineer uses a public API to collect 1m bsky posts and shares them via HF. the retaliation from the community is huge, with many people claiming a GDPR violation. i wonder if anybody from privacy law bluesky has an opinion on what happened. #privacy #law #gdpr

data catalog = select * from INFORMATION_SCHEMA.Tables; data lineage = orchestration semantic layer = (modeling) metrics and KPIs

Is Trump 2.0 like the Cultural Revolution? Let's hope not: foreignpolicy.com/2024/11/27/t...

sassy ai monitoring

1. Can you stop companies from training generative AI using your data? No, not currently. 2. Is this dataset meant for training generative AI? 🤷‍♀️ but more likely for research and statistical analysis. 3. Is it ok to duplicate and distribute people’s data without agency to opt out? I’d argue no.